Statistical modeling and performance characterization of a real-time dual camera surveillance system

Abstract

The present invention relates to a method for visually detecting and tracking an object through a space. The method chooses modules for a restricting a search function within the space to regions with a high probability of significant change, the search function operating on images supplied by a camera. The method also derives statistical models for errors, including quantifying an indexing step performed by an indexing module, and tuning system parameters. Further the method applies a likelihood model for candidate hypothesis evaluation and object parameters estimation for locating the object.

1 . A surveillance method, comprising: 
 determining a location of an object with a first camera;    determining an uncertainty of the location of the object; and    controlling a characteristic of a second camera in accordance with the uncertainty.    
   
   
       2 . The method of  claim 1 , wherein the first camera includes a camera and a parabolic mirror and the second camera is a pan-tilt camera.  
   
   
       3 . The method of  claim 1 , wherein the characteristic of the second camera is zoom.  
   
   
       4 . The method of  claim 1 , further comprising controlling one or more additional characteristics of the second camera in accordance with the uncertainty.  
   
   
       5 . The method of  claim 4 , wherein the first characteristic is zoom and the one or more characteristics are pan and tilt.  
   
   
       6 . The method of  claim 1 , wherein the characteristic of the second camera is also controlled in accordance with the location.  
   
   
       7 . The method of  claim 3 , wherein the second camera zooms in as uncertainty decreases.  
   
   
       8 . The method of  claim 5 , wherein the object is a person's face and the controlling step is performed so that a probability that the person's face is captured by the second camera is within a user specified probability.  
   
   
       9 . A system for monitoring an object, comprising: 
 a first camera;    a second camera;    a processor that determines an uncertainty of a location of the object and controls a characteristic of the second camera in accordance with the uncertainty.    
   
   
       10 . The system of  claim 9 , wherein the first camera includes a camera and a parabolic mirror and the second camera is a pan-tilt camera.  
   
   
       11 . The system of  claim 9 , wherein the characteristic of the second camera is zoom.  
   
   
       12 . The method of  claim 9 , further comprising controlling one or more additional characteristics of the second camera in accordance with the uncertainty.  
   
   
       13 . The method of  claim 12 , wherein the first characteristic is zoom and the one or more characteristics are pan and tilt.  
   
   
       14 . The system of  claim 9 , wherein the characteristic of the second camera is also controlled in accordance with the location.  
   
   
       15 . The system of  claim 11 , wherein the second camera zooms in as uncertainty decreases.  
   
   
       16 . The method of  claim 9 , wherein the object is a person's face and the controlling step is performed so that a probability that the person's face is captured by the second camera is within a user specified probability.  
   
   
       17 . A system for tracking and detecting objects, comprising: 
 an automobile; and    an omnidirectional camera mounted in the automobile.    
   
   
       18 . The system of  claim 17 , further comprising a parabolic mirror mounted in the automobile that interfaces with the omnidirectional camera to provide a panoramic view.  
   
   
       19 . The system of  claim 18 , further comprising a pan-tilt camera mounted in the automobile.  
   
   
       20 . The system of  claim 19 , wherein the pan-tilt camera is controlled in accordance with an output of the omnidirectional camera.  
   
   
       21 . The system of  claim 20 , wherein a zoom characteristic of the pan-tilt camera is controlled.  
   
   
       22 . A method for locating an object near an automobile, comprising: 
 creating a panoramic view with a parabolic mirror located in the automobile; and    capturing the panoramic view from the parabolic mirror with an omnidirectional camera located in the automobile.    
   
   
       23 . The method of  claim 22 , further comprising processing the panoramic view to determine a location of the object.  
   
   
       24 . The method of  claim 23 , further comprising controlling a pan-tilt camera based on the location of the object.  
   
   
       25 . The method of  claim 24 , further comprising controlling a zoom characteristic of a pan-tilt camera based on the location of the object.  
   
   
       26 . The method of  claim 25 , further comprising determining an uncertainty of the location and controlling a zoom characteristic of a pan-tilt camera based on the uncertainty.  
   
   
       27 . A method of locating an object, comprising: 
 defining a plurality of statistical models of a scene;    defining one or more tasks to be performed;    translating the plurality of statistical models and the one or more tasks to specify one or more index functions that generate one or more hypotheses of object location candidates;    optimizing one or more parameters of each of the one or more index functions; and    creating an estimate that will follow use of the one or more index functions to verify the one or more hypotheses and to estimate object parameters.    
   
   
       28 . The method of  claim 27 , wherein the estimate is used to generate uncertainty estimates.  
   
   
       29 . The method of  claim 28 , comprising controlling a pan-tilt-zoom of a camera based on the uncertainty estimates.  
   
   
       30 . The method of  claim 29 , comprising setting parameters of a sensor system based on the analysis of the behavior of the one or more index functions.  
   
   
       31 . The method of  claim 27 , wherein the one or more tasks includes object detection, object localization and/or object tracking.  
   
   
       32 . The method of  claim 27 , wherein the plurality of models are selected from the group consisting of: a calibration model, an illumination invariant model, a background adaptation model, an object geometry model, a camera geometry model, an error model and an illumination model.